🧔♂️ A friendly human may check it before it goes live. More news here
CoreWeave shares slip on weak 2025 revenue forecast
CoreWeave shares fell 6% in after-hours trading after issuing full-year revenue guidance below analyst expectations, despite strong Q3 results.
The AI infrastructure provider, which rents out Nvidia GPUs, reported Q3 revenue of US$1.4 billion, beating the US$1.3 billion estimate and up 134% year-on-year.
Net loss narrowed to US$110 million from US$360 million a year ago.
The company now expects 2025 revenue between US$5.1 billion and US$5.2 billion, below the US$5.3 billion consensus.
CEO Mike Intrator said a third-party data center delay will not impact its US$55.6 billion backlog.
Recent deals include a US$6.5 billion expansion with OpenAI, a six-year agreement with Meta worth up to US$14.2 billion, and a new contract with a large cloud provider.
Shares, up 164% since its March IPO, slipped after the earnings report.
🔗 Source: CNBC
🧠 Food for thought
Implications, context, and why it matters.
CoreWeave’s $55.6B backlog raises timing and quality questions
- The backlog rose to $55.6B from $15.1B in its S-1 (the U.S. IPO registration filing) 1, with OpenAI’s $22.4B and Meta’s $14.2B deals 2. It has not said how much is take-or-pay (the buyer pays regardless of usage) versus cancellable, or when cash will come in.
- Earlier it said about 50% (~$15B) would turn into revenue within 24 months 2. Most of the rest sits further out, so the headline number says little about near-term results.
- Management reported a third-party data center delay, yet backlog stays intact. Annual capital expenditures (capex) of $20–23B 2 and dependence on outside sites can push revenue out even if demand holds.
AI teams can benefit from tight GPU supply
- Only about 20% of its contracted 2.2 GW, or 2,200 MW, power was active in Q2 2. A recent delay makes clear that even specialized GPU providers hit infrastructure bottlenecks.
- Teams building FinOps (cloud financial operations) tools should create orchestration that compares price plus service-level agreements across CoreWeave, hyperscalers (the largest public cloud platforms) and neocloud providers (smaller, specialized clouds). Track preemption rates (how often lower-priority instances are interrupted) and egress fees (costs to move data out) to improve cost and uptime.
- CoreWeave reports 35–45% Model floating-point operations per second (FLOPS) utilization 2. Providers deliver different results. Teams that measure training speed plus cost can gain 25–50% efficiency by picking the right vendor for each job.
